Nvidia Tegra CPU: "Tick, Tock" strategy?

[cbn]

http://www.anandtech.com/show/5118/qualcomm-announces-a-bunch-of-krait-based-snapdragon-s4-socs

When I saw this Qualcomm SOC product announcement, I thought to myself "Oh man, have they really missed their clock targets!"

Back when Qualcomm originally announced "Krait", up to 2.5 Ghz was mentioned for the quad core!

Surely the higher clocked Kraits will come later, but this really got me thinking if Nvidia's SOC roadmap was planned on purpose to avoid the problem of "new architecture on new process".

With Nvidia releasing Tegra 3 on 40nm (so late in 2011) it has delayed their ARM A15 offering to a launch point well after the competition. To compensate Nvidia will release Tegra 3 (an old design) on 28nm (new node) around the same time as the first Cortex A15s launch (new design on new node).

In the 2H of 2012:

Nvidia Tegra 3 on 28nm

Vs.

Competition's First Cortex A15 on 28nm


In 2013

Nvidia First Cortex A15

vs.

Competition's "re-spin" Cortex A15

Just wondering if Nvidia operating on "Tick Tock" will help or hurt their overall strategy? Will Tegra 3 on 28nm give them enough experience with the node to launch an A15 competitive with other Cortex A15 competition in 2013?

 

[happy medium]

Just wondering if Nvidia operating on "Tick Tock" will help or hurt their overall strategy?

No, I think the just released 5 core tegra 3 is the latest and greatest.
The fastest SOC with the best battery life.

The next shrunk 28nm tegra 3 will give higher clocks with more battery life.

I think they are doing the right thing, they are still ahead of the competition by 8 to 10 months.

The 5th core and battery life is the key to there success IMHO.

 

[cbn]

The next shrunk 28nm tegra 3 will give higher clocks with more battery life.

It will be interesting to see how high Nvidia will go with those Cortex A9 clocks on 28nm.

Nvidia A9= 4 small cores clocked high?

vs.

Cortex A15= 2 larger cores clocked lower?

 

[ocre]

The competition's claims may not pan out. I will bet money on them being delayed. The 28nm node has been tough. To think they have a new node and process, its gonna be really hard to have shipping parts so soon. I think nvidia is and will work at an extremely fast pace so as not to be left behind.

too many ppl are like, "nvidias tegra3 is dumb cause its gonna get blown away by a15 cores in 6months"

Thats crazy to think. ARM is now in a race like there has never been. Nvidia is now the hottest ARM chip out, and this was smart. By the time the A15s come out, nvidia wont be standing still. Its gonna be tough for these companies to get these chips shipping in 6months. Nvidia made a clever move. They are very serious, their strategy is a fierce one. They arent gonna be left behind, i really cannot see this happening.

Aggressive as they possibly can be.

 

[cbn]

ARM is now in a race like there has never been.

Yep, it is amazing how the first ARM to land 45nm came middle of 2010. (Previous to that were were looking at Cortex A8s on 65nm as late as early 2010!)

But now look how fast they are to accept current nodes.

IMO Cortex A9 was the turning point for ARM.

 

[AtenRa]

http://www.anandtech.com/show/5118/qualcomm-announces-a-bunch-of-krait-based-snapdragon-s4-socs

When I saw this Qualcomm SOC product announcement, I thought to myself "Oh man, have they really missed their clock targets!"

Back when Qualcomm originally announced "Krait", up to 2.5 Ghz was mentioned for the quad core!

Cortex A15 target was 1.5GHz at 32/28nm LP process and 2.5GHz at 32/28nm HP process.

 

[Cerb]

too many ppl are like, "nvidias tegra3 is dumb cause its gonna get blown away by a15 cores in 6months"

While those are still undergoing last minute testing by manufacturers, and in the US, by wireless carriers, NV will be putting money in the bank from sales that have been reaching consumers and increasing their mindshare (I hate such marketing terms, but it is the right word), and gaining a reputation for meeting advertised feature and performance targets in a timely manner.

 

[cbn]

Cortex A15 target was 1.5GHz at 32/28nm LP process and 2.5GHz at 32/28nm HP process.

Not sure how the Custom Krait CPU compares to the stock Cortex A15. (I'll have to read up on this later)

All I know is that APQ8064 came in lower than I expected:

Although it's not new, the APQ8064 is also on the roadmap. The 8064 is a quad-core Krait SoC, clock speeds are once again limited to 1.7GHz initially (the SoC will eventually scale to much higher speeds). The memory interface grows to support standard DDR3-1066. The SoC also has on-die SATA and PCIe controllers, making it an obvious target for Windows 8 based systems.

http://www.qualcomm.com/news/releas...d-core-snapdragon-next-generation-tablets-and

Qualcomm Announces Quad-Core Snapdragon for Next Generation Tablets and Computing Devices
New Chipset Includes Next Generation CPU Architecture with Quad-Core and Adreno GPU


Barcelona, Spain – February 14, 2011 – Qualcomm Incorporated (NASDAQ: QCOM) today announced its quad-core Snapdragon™ chipset designed to meet the requirements of next generation tablets and computing devices. The new quad-core APQ8064™ is the flagship chipset in the new family of Snapdragon chipsets and is based on the new micro-architecture code named “Krait.” With the purpose of being built for mobile devices, this 28nm micro-architecture will redefine performance, achieving speeds of up to 2.5GHz per core and minimizing power consumption and heat generation to enable new, thin and light form factors.

The Snapdragon APQ8064 chip will be designed to enable the next generation of converged computing and entertainment devices. These devices will have significantly higher performance requirements, including support for larger screen sizes and resolutions, more complex operating systems, multi-tasking, multi-channel audio, HD gaming and stereoscopic 3D (S3D) photo and video capture and playback, as well as output in full HD to 1080P flat panel displays over HDMI.

While performance requirements have been increasing, battery technology and capacity have struggled to develop at the same pace. To meet this challenge, Qualcomm created its next generation architecture and integrated four new, low-power CPU cores and its advanced Adreno® graphics into the APQ8064, enabling it to offer twelve times the available performance as well as 75 percent lower power than the first generation of Snapdragon processors. The combination of advanced processors and multimedia technology will provide tablets and mobile computing devices with unsurpassed performance, battery life, low thermal dissipation and the broadest set of connectivity options available in the industry.

“We’re excited about the new, innovative devices the APQ8064 will enable in the market,” said Luis Pineda, senior vice president of computing and consumer products at Qualcomm.
“With its WiFi integration and ability to seamlessly interface with Qualcomm’s 3G and LTE modules, the APQ8064 will provide OEM’s with a flexible, cost efficient and fast time to market platform that can meet all of their design configuration needs for tablets and next generation computing and consumer electronic devices.”

The APQ8064 processor will include the Adreno 320 quad-core GPU, which will deliver fifteen times greater performance than the original Adreno GPU, for a console-quality gaming experience, as well as rendering rich UI’s. With up to 20 Megapixel camera support, the APQ8064 will internally synchronize two camera sensors for 3D video recording and will support external 3D video playback.

The APQ8064 will feature a variety of features that make it the most compelling processing solution for use in mobile entertainment and computing devices, including support for both PCDDR3 and LPDDR2 memory, serial and PCIe interfaces and multiple USB ports.

As part of the next generation Snapdragon family, the APQ8064 will share a core set of features, including a quad-combo of connectivity solutions (i.e. WLAN, GPS, BT and FM), support for NFC and stereoscopic 3D (S3D) video and photo capture and playback. The APQ8064 also will support all available major operating systems and be software compatible across the chipset family.

Samples of the APQ8064 are anticipated to be available in early 2012.

 

[cbn]

http://www.anandtech.com/show/4940/qualcomm-new-snapdragon-s4-msm8960-krait-architecture

Digging through the Anandtech Krait Architecture Article, I was surprised to see such an early release date for Qualcomm.

Qualcomm Krait (28nm): 1H 2012
TI OMAP 5 (28nm): 2H 2012
Everyone else's Cortex A15 (28nm): 2013

For some reason I had thought Krait was a 2H 2012 release along with other Cortex A15s.

 

[ocre]

There is no way anyone will see Krait in devices 1H 2012. If so we wouldnt have heard Qualcomm talking smack on nvidias tegra3. They are going public to try to slow the tegra hype cause they are currently left behind. Also even when it does come out, i dont think they will reach those speeds for a while. The tegra3 is a beast and Qualcomm is threaten. Now all they got is, "uhm jut wait....are custom design is better.....um nvidia didnt out class us.....uhm"

Qualcomm will eventually get their chip out, and it will be nice. I would bet their first gen krait will be under clocked but still be close to the tegra3 performance. I also doubt we will see anything like krait till 2H. By then nvidia will die shrink their tegra3 which will also be a beast.

Nvidia is also designing custom cores. Just by looking at their tegra work, i would bet their fully custom ARM chips will be out of this world. You dont even have to like nvidia to be impressed with their ARM entry. Impressive is such an understatement. in just a few months they have dominated. They are taking ARM to an all time high. They have the petal to the floor and its no stopping for anyone!!!!

 

[Dribble]

With Nvidia releasing Tegra 3 on 40nm (so late in 2011) it has delayed their ARM A15 offering to a launch point well after the competition.

How do you know that nvidia will release A15 after the competition? Tegra 2 and 3 have released well before the competition, why do you think that will change for tegra 4?

Nvidia just released an intermediate chip which no competitor has an answer too. The only reason the competitors are singing about A15's is because that's all they can do despite it not being due out for quite a while yet. Just because nvidia isn't discussing their A15 chips much right now doesn't mean they have taken their foot of the throttle. When the A15's finally make it to market chances are nvidia will have one of the first chips to use it.

 

[Idontcare]

There is no way anyone will see Krait in devices 1H 2012. If so we wouldnt have heard Qualcomm talking smack on nvidias tegra3.

This is not uncommon in this industry. Schedule slips are the norm for everyone involved because everyone involved is using the same foundry (TSMC) and that means there is no real way to release ahead of your competitor unless your competitor screws up (needs a respin).

But what more typically happens is the foundry screws up and misses their schedule, pushing everyone else back by quarters at a time.

Nvidia knows this, hence they stuck with 40nm. If you think about it, they are kinda counting on TSMC being TSMC and screwing the customers with 28nm delays, which will play to Nvidia's advantage. (and who best to know TSMC than one of their highest volume customers for leading edge nodes?)

 

[cbn]

Maybe there is an economic reason for Qualcomm's ability to launch this new SOC earlier than the competition:

With Qualcomm's market share and chip volumes, I just wonder if their architectural license gives them an advantage to launch products on TSMC's more expensive wafers? (ie, Qualcomm pays more for wafers, but saves on core licensing costs from ARM)

 

[ocre]

This is not uncommon in this industry. Schedule slips are the norm for everyone involved because everyone involved is using the same foundry (TSMC) and that means there is no real way to release ahead of your competitor unless your competitor screws up (needs a respin).

But what more typically happens is the foundry screws up and misses their schedule, pushing everyone else back by quarters at a time.

Nvidia knows this, hence they stuck with 40nm. If you think about it, they are kinda counting on TSMC being TSMC and screwing the customers with 28nm delays, which will play to Nvidia's advantage. (and who best to know TSMC than one of their highest volume customers for leading edge nodes?)

Goes very well with my point.

 

[ocre]

Maybe there is an economic reason for Qualcomm's ability to launch this new SOC earlier than the competition:

With Qualcomm's market share and chip volumes, I just wonder if their architectural license gives them an advantage to launch products on TSMC's more expensive wafers? (ie, Qualcomm pays more for wafers, but saves on core licensing costs from ARM)

Wait till they launch. Then we can talk about it.

Its common practice to co@k block when you cant compete. Qualcomm is wanting to dampen the tegra3 launch in order to dampen its sales. They want to prevent loosing market by convincing OEMs that tegra3 is not all that and they will soon be eating it for lunch. The more OEMs they get to wait the better. This is the only reason they have even spoken out against tegra3. They dont have to be ready by H1 for this to work. Its all about keeping the market.

 

[Idontcare]

Maybe there is an economic reason for Qualcomm's ability to launch this new SOC earlier than the competition:

With Qualcomm's market share and chip volumes, I just wonder if their architectural license gives them an advantage to launch products on TSMC's more expensive wafers? (ie, Qualcomm pays more for wafers, but saves on core licensing costs from ARM)

You are talking the opportunity of saving pennies by spending dollars. ARM's cut just isn't that much for it to make a big difference if they reduced it by 10%.

Qualcomm commands their market volume because of the IP they add to the equation, all the non-ARM stuff that goes into their SOC.

 

[cbn]

You are talking the opportunity of saving pennies by spending dollars. ARM's cut just isn't that much for it to make a big difference if they reduced it by 10%.

Qualcomm commands their market volume because of the IP they add to the equation, all the non-ARM stuff that goes into their SOC.

Yes, but look at how Qualcomm is selling their SOCs:

MSM8960: Two 1.7 Ghz Krait cores, (Initial speeds will be 1.5 Ghz), Adreno 225, dual channel LPDDR2, 3G/4G/LTE baseband on-die, 20 MP camera

MSM8930: Two 1.2 Ghz Krait cores, Adreno 305, single channel LPDDR2, integrated LTE, 13.5 MP camera

Clearly, but comparing feature sets MSM8960 and MSM8930 are two different chips, not merely binned versions of each other. Therefore my conclusion is that Qualcomm is downclocking perfectly good cpus in the MSM8930 series in order to mark up their more expensive MSM8960 SOCs (containing the faster CPU)

Maybe a caveat would be the cost of the dual channel memory controller vs the cost of a the single channel memory controller. If the 1.7 Ghz CPU demands dual channel (and this memory controller is expensive) that might give Qualcomm enough cause to downclock the CPUs in MSM8930.

 

[cbn]

Qualcomm commands their market volume because of the IP they add to the equation, all the non-ARM stuff that goes into their SOC.

The major non-ARM IP would be the integrated baseband?

This is so much more effective over custom CPU integration because of power savings? Or other reasons?

Anyone?

 

[Cerb]

The major non-ARM IP would be the integrated baseband?

That, Adreno, DSP, and whatever gives them such good memory performance (their cheap chips are generally faster than others' cheap chips). Also, sufficient software support, which many a hardware vendor ignores, IBM-style ("You want quality, well-tested, drivers for a new kernel version? Really? Everybody else just flails around in the dark until it works, and seems happy."). NVidia also gets that they need to sell a complete platform, not just one of several chunks of silicon that sits behind a display.

 

[soccerballtux]

No, I think the just released 5 core tegra 3 is the latest and greatest.
The fastest SOC with the best battery life.

The next shrunk 28nm tegra 3 will give higher clocks with more battery life.

I think they are doing the right thing, they are still ahead of the competition by 8 to 10 months.

The 5th core and battery life is the key to there success IMHO.

am seriously considering dumping some $$$ into NVDA long term.

 

[soccerballtux]

Nvidia knows this, hence they stuck with 40nm. If you think about it, they are kinda counting on TSMC being TSMC and screwing the customers with 28nm delays, which will play to Nvidia's advantage. (and who best to know TSMC than one of their highest volume customers for leading edge nodes?)

ha interesting...

 

[cbn]

ARM is now in a race like there has never been.

The race has been very interesting to follow.

Here is a great chart posted by IdontCare in another thread:

With Pollack's rule and ARM running out of process nodes to advance on, I just wonder what the next step will be?

Wider ARM cpus in smartphone chassis vs. increasing ARM cpu core count vs. increasing integration of non-cpu components within the SOC?

 

[soccerballtux]

With Pollack's rule and ARM running out of process nodes to advance on, I just wonder what the next step will be?

I don't think it will matter. At that point they will be fast enough for facebook, internet browsing, and streaming music and videos; which is all anybody (except gamers) really uses their computer for anyways...

 

[soccerballtux]

The potential virtualization holds with ARMv8 is exciting, however. Imagine spinning off the first 15MB of all of your Android apps' RAM usage (the Java backend/VM) into the same virtual pages. That would free up tons of RAM per app...not that we'll need more free RAM, but it'll be cool nonetheless.

 

[cbn]

I don't think it will matter. At that point they will be fast enough for facebook, internet browsing, and streaming music and videos; which is all anybody (except gamers) really uses their computer for anyways...

I was just thinking about how much performance gain someone with a 1H 2010 smartphone would see upon upgrading (through 2 year contract perk) to a 1H 2012 smartphone:

1H 2010 Qualcomm: 65nm Single core scorpion @ 1000 Mhz
1H 2012 Qualcomm: 28nm Dual core Krait @ 1.5 Ghz

Wow! That is a huge jump in CPU power!

Maybe the next selling point will be battery life (if the ARM CPU race starts to slow down)? Maybe even heterogeneous computing (Eight core ARM Mali T658 with 272 Gflops, etc)?